Effective professional communication on platforms such as LinkedIn increasingly demands the ability to adapt message tone and style to suit different audiences and contexts. Whether the goal is to convey approachability through a casual tone or to demonstrate professionalism through formality, the capacity to flexibly transform written communication is invaluable. However, manual rewriting is subjective, inconsistent, and impractical at scale. This thesis addresses these challenges by proposing a fully automated, reinforcement learning-based system for bidirectional text style transfer, transforming messages from formal to casual and vice versa. The core of this research is the application and extension of the RLPrompt framework, a state-of-the-art method that casts text style transfer as a discrete prompt optimization problem.